Reasoning with a predicate inference network

Presents a formalism for commonsense reasoning with variables using an inference network. It is a three-layer acyclic network that is based on a four truth-value interpretation (true, false, contradictory and undefined) with degree-of-belief value. The knowledge is segregated into objects, predicates and variables. Inter-layer links are used to relate objects with their respective predicate and the instantiations of rules and other relations during reasoning. The main knowledge is kept in the predicate layer as nodes and links. Typical relations are logical relations such as AND, OR and NOT, IF-THEN rules and human-biased relations. Reasoning in this network is viewed as information propagation amongst the instantiated predicate nodes via various directed links. This knowledge system accepts inconsistent knowledge declarations and local inconsistency reasoning is built into the inferencing processes. The proposed information source checking algorithm, not only extends the recursive rules that can be controlled locally, but they also always terminate deterministically